Design challenges and solutions for ultra-high-density monolithic 3D ICs

Monolithic 3D IC is an emerging technology to continuously satisfy demands for power reduction under challenges posed by traditional device scaling. In this paper, for the first time, we study power benefits of 4-tier monolithic 3D ICs compared with 2-tier monolithic 3D and 2D ICs. We present a tier partitioning methodology that significantly extends the capability of a state-of-the-art flow built for 2-tier monolithic 3D ICs. We develop two complete RTL-to-GDSII design flows to achieve this goal and offer quantitative comparisons. In addition, we study impacts of inter-tier via usage on 2-tier and 4-tier monolithic 3D ICs. Our experiments show that poorly controlled inter-tier via usage results in up to 6.05% degradation in total power savings. Thus, we develop an effective strategy to achieve inter-tier via configurations to optimize power metrics. Experiments show that 4-tier monolithic 3D ICs outperform 2-tier and 2D IC by 15% and 50% in terms of power and 25% and 75% in area under the same performance.

The state-of-the-art Monolithic 3D (M3D) IC design methodologies~\citem3d:Ku-tcad-Compact2D, m3d:Panth-tcad-Shrunk2D use commercial electronic design automation tools built for 2D ICs to implement a pseudo-3D design and split it into two dies that are routed independently to create an M3D design. Therefore, an accurate estimation of 3D wire parasitics at the pseudo-3D stage is important to achieve a well optimized M3D design. In this paper, we present a regression model based on boosted decision tree learning to better predict the 3D wire parasitics (RCs) at the pseudo-3D stage. Our model is trained using individual net features as well as the full-chip design metrics using multiple instantiations of 8 different netlists and is tested on 3 unseen netlists. Compared to the Compact-2D~\citem3d:Ku-tcad-Compact2D flow on its own as the reference pseudo-3D, the addition of our predictive model achieves up to $2.9 \times$ and $1.7 \times$ smaller root mean square error in the resistance and capacitance predictions respectively. On an unseen netlist design, we observe that our model provides 98.6% and 94.6% RC prediction accuracy in 3D and up to $6.4 \times$ smaller total negative slack of the design compared to the result of Compact-2D flow resulting in a more timing-robust M3D IC. This model is not limited to Compact-2D, and can be extended to other pseudo-3D flows.

Network-on-chips (NoCs) with traditional regular topologies generally have an average-hop-count increased, polynomially, with the network size. This brief proposes to generate brain-network-inspired (BNI) topologies on monolithic 3D (M3D) ICs, where the average hop count scales up logarithmically, for large-scale systems with tens of thousands of cores. We propose a routability-driven method to generate BNI topologies on M3D ICs with scale-free and small-world properties. Moreover, we present an iterative method to reduce the network diameter using as few tier-crossing routers as possible. The resulting BNI topologies not only have over 50% lower average hop count and lower diameter than traditional regular topologies, but also have shorter total link length than other irregular topologies. The performance is further validated by the cycle-accurate simulation.

Monolithic inter-layer vias (MIVs) used in 3D integration technology are much smaller than through-silicon vias (TSVs) used in TSV-based 3D integration. Thus, monolithic 3D integration provides shorter wire length, better performance, and lower power consumption than TSV-based 3D integration. Multi-tier monolithic 3D integration stacks more than two silicon layers, thereby increasing the device density, which reduces the wire length and power consumption and improves performance further than two-tier monolithic 3D integration. Since the size of an MIV is comparable to that of a local via, the wire lengths of multi-tier gate-level monolithic 3D IC layouts are expected to be close to the ideal wire length reduction ratio, for example, 29, 42, and 50 percent for two-, three-, and four-tier designs, respectively. However, it is still unknown whether it would be possible to achieve the ideal wire length reduction ratios by multi-tier gate-level monolithic 3D integration. In this paper, we propose an efficient routing methodology to design multi-tier monolithic 3D ICs. Using the design methodology, we design two- to eight-tier gate-level monolithic 3D IC layouts and thoroughly investigate the wire length characteristics of the layouts. The simulation results show that the proposed routing methodology reduces the wire length by 12 to 53 percent compared to the 2D designs. In addition, the proposed routing methodology builds multi-tier monolithic 3D IC layouts more efficiently than a state-of-the-art methodology with comparable routed wire length.

Monolithic 3D ICs (M3D) are fabricated using a sequential process that grows new device and interconnect tiers in a bottom-up fashion. This fabrication process is in contrast to through-silicon via (TSV) technology that aligns and bonds pre-built tiers. M3D offers key advantages over TSVs, including (1) orders-of-magnitude smaller inter-tier vias, (2) no need for high alignment accuracy, (3) finer-grained tier partitioning options, etc. Recent studies have shown power, performance, area, and reliability (PPAR) advantages of M3D over TSV. However, M3D also suffers from its own problems, including (1) device and interconnect performance mismatch between tiers, (2) lack of EDA solutions, (3) testing challenges, (4) cost, etc. Research efforts have also been made to model and mitigate the impact of these undesirable characteristics of M3D. We will provide a survey of work that address the above issues and conclude with future directions.

Monolithic 3D (M3D) is an emerging technology that enables integration density which is orders of magnitude higher than that offered by through-silicon-vias. In this paper, we demonstrate that a modified 2D placement technique coupled with a post-placement partitioning step is sufficient to produce high-quality M3D placement solutions. We also present a commercial router-based monolithic intertier via insertion methodology that improves the routability of M3D ICs. We demonstrate that, unlike in 2D ICs, the routing supply and demand in M3D ICs are not completely independent of each other. We develop a routing demand model for M3D ICs, and use it to develop an O(N) min-overflow partitioner that enhances routability by off-loading demand from one tier to another. This technique reduces the routed wirelength and the power delay product by up to 7.44% and 4.31%, respectively. This allows a two-tier M3D IC to achieve, on average, 19.9% and 11.8% improvement in routed wirelength and power delay product over 2D, even with reduced metal layer usage.

Resistive random-access memory (RRAM) constitutes a promising technology for next-generation memory architectures due to its simple structure, high on/off ratio, and processing-in-memory ability. Its compatibility with emerging monolithic 3D (M3D) integration enables extremely high density using monolithic inter-tier vias (MIVs). However, both RRAM and M3D are susceptible to high defect rates due to immature manufacturing processes and process variations. Fault diagnosis for M3D-integrated RRAM and MIVs is therefore necessary to facilitate yield learning. In this work, we present a detailed characterization of RRAM faulty behaviors in the presence of process variations and manufacturing defects. We develop a diagnosis procedure by identifying appropriate reference resistance based on RRAM characteristics to efficiently distinguish fault origins. Results show that the proposed solution is compatible with existing test algorithms to significantly improve diagnostic resolution without affecting fault coverage.

Monolithic 3D IC overcomes the limitation of the existing through-silicon-via (TSV) based 3D IC by providing denser vertical connections with nano-scale inter-layer vias (ILVs). In this paper, we demonstrate a thorough RTL-to-GDS design flow for monolithic 3D IC, which is based on commercial 2D place-and-route (P&R) tools and clever ways to extend them to handle 3D IC designs and simulations. We also provide a low-cost built-in-self-test (BIST) method to detect various faults that can occur on ILVs. Lastly, we present a resistive random access memory (ReRAM) compiler that generates memory modules that are to be integrated in monolithic 3D ICs.

In this paper, we present an optimization flow for monolithic 3D ICs called Pin-3D Optimizer. Compared with the state-of-the-art RTL-to-GDS flows that rely on ad-hoc technology file tweaks and RC scaling, Pin-3D offers a streamlined method to run commercial 2D IC tools to obtain high-quality monolithic 3D IC designs. Specifically, Pin-3D supports effective legalization, routing, timing closure, and ECO optimization for monolithic 3D IC designs. We propose a novel optimization methodology where the cells in each tier of a 3D IC are optimized using cell data and constraints of the full 3D design. The optimizations in a tier also directly influence the timing, power in the other tiers, leading to better overall PPA of the 3D IC. With the help of two industry processors designed with a 28 nm technology node, we show that Pin-3D provides up-to 9.0% smaller wirelength and 88% smaller total negative slack than die-by-die M3D flows. We also observe up-to 8.7% lower power and 26% smaller wirelength than 2D ICs. In addition, Pin-3D is the first flow that supports routing and timing optimization in heterogeneous logic-on-logic monolithic 3D ICs. We demonstrate this capability by performing area-balanced tier partitioning, routing, and timing closure of a 3D design with different technologies on each die.

Monolithic three-dimensional (3D) integration enables the most fine-grained integration of transistors by stacking very thin layers and fabricating monolithic inter-layer vias as small as local vias. Thus, monolithic 3D integration is expected to provide a higher degree of wirelength reduction, performance improvement, and power saving. Due to the prospective properties of the monolithic 3D integration technology, research on multilayer monolithic 3D integration that stacks more than two device layers is also ongoing. In this paper, we propose an algorithm that optimizes dynamic power consumption of gate-level monolithic 3D ICs. Under the same timing constraints, our algorithm reduces dynamic power consumption more effectively than a uniform-scaling-based placement algorithm. We also design multi-tier monolithic 3D ICs and show that our algorithm outperforms the uniform-scaling-based placement algorithm by 11.4% on average.

Monolithic inter-tier vias (MIVs) in monolithic 3-D integrated circuits (M3D ICs) enables massive vertical integration. However, MIVs are more susceptible to defects due to high integration density and complex manufacturing processes. Existing MIV test techniques can effectively detect and locate MIV faults, but diagnosable fault types are limited. We propose a novel fault diagnosis method based on signal transmission performance analysis. This method can diagnose more fault types, including resistive open, hard open, short, and leakage faults. In the proposed solution, fault diagnosis can be carried out by comprehensively monitoring voltage and delay characteristics of MIVs. The effectiveness of fault diagnosis is verified through high speed simulation program with integrated circuit emphasis (HSPICE) simulations. We also perform Monte Carlo simulations to prove that the proposed method has high robustness even in the case of process variations. Experimental results show that the proposed method has low hardware overhead while ensuring high diagnostic resolution.

Heterogeneous manycore architectures are the key to efficiently execute compute- and data-intensive applications. Through-silicon-via (TSV)-based 3D manycore system is a promising solution in this direction as it enables the integration of disparate computing cores on a single system. Recent industry trends show the viability of 3D integration in real products (e.g., Intel Lakefield SoC Architecture, the AMD Radeon R9 Fury X graphics card, and Xilinx Virtex-7 2000T/H580T, etc.). However, the achievable performance of conventional TSV-based 3D systems is ultimately bottlenecked by the horizontal wires (wires in each planar die). Moreover, current TSV 3D architectures suffer from thermal limitations. Hence, TSV-based architectures do not realize the full potential of 3D integration. Monolithic 3D (M3D) integration, a breakthrough technology to achieve “More Moore and More Than Moore,” opens up the possibility of designing cores and associated network routers using multiple layers by utilizing monolithic inter-tier vias (MIVs) and hence, reducing the effective wire length. Compared to TSV-based 3D integrated circuits (ICs), M3D offers the “true” benefits of vertical dimension for system integration: the size of an MIV used in M3D is over 100 × smaller than a TSV. This dramatic reduction in via size and the resulting increase in density opens up numerous opportunities for design optimizations in 3D manycore systems: designers can use up to millions of MIVs for ultra-fine-grained 3D optimization, where individual cores and routers can be spread across multiple tiers for extreme power and performance optimization. In this work, we demonstrate how M3D-enabled vertical core and uncore elements offer significant performance and thermal improvements in manycore heterogeneous architectures compared to its TSV-based counterpart. To overcome the difficult optimization challenges due to the large design space and complex interactions among the heterogeneous components (CPU, GPU, Last Level Cache, etc.) in a M3D-based manycore chip, we leverage novel design-space exploration algorithms to trade off different objectives. The proposed M3D-enabled heterogeneous architecture, called HeM3D , outperforms its state-of-the-art TSV-equivalent counterpart by up to 18.3% in execution time while being up to 19°C cooler.

Monolithic 3D (M3D) technology enables unprecedented degrees of integration on a single chip. The miniscule monolithic inter-tier vias (MIVs) in M3D are the key behind higher transistor density and more flexibility in designing circuits compared to conventional through silicon via (TSV)-based architectures. This results in significant performance and energyefficiency improvements in M3D-based systems. Moreover, the thin inter-layer dielectric (ILD) used in M3D provides better thermal conductivity compared to TSV-based solutions and eliminates the possibility of thermal hotspots. However, the fabrication of M3D circuits still suffers from several non-ideal effects. The thin ILD layer may cause electrostatic coupling between tiers. Furthermore, the low-temperature annealing degrades the top-tier transistors and bottom-tier interconnects. An NoC-based manycore design needs to consider all these M3D- process related non-idealities. In this paper, we discuss various design challenges for an M3D-enabled manycore chip. We present the power-performance-thermal trade-offs associated with these emerging manycore architectures.

Deep neural network (DNN) models are being expanded to a broader range of applications. The computational capability of traditional hardware platforms cannot accommodate the growth of model complexity. Among recent technologies to accelerate DNN, resistive memory (ReRAM)-based processing-in-memory (PIM) emerged as a promising solution for DNN inference due to its high efficiency for matrix-based computation. We face two major technical challenges in extending the use of ReRAM-based accelerators for training: (1) full-precision data is essential in back-propagation; (2) the need to support both feed-forward and back-propagation aggravates the data-movement burden. We propose a heterogeneous architecture named as 3D-ReG, which leverages full-precision GPU to ensure training accuracy and low-overhead 3D integration to provide low-cost data movements. Moreover, we introduce conservative and aggressive task-mapping schemes, which partition the computation phases in different ways to balance execution efficiency and training accuracy. We evaluate 3D-ReG implemented with two 3D integration technologies, through-silicon vias (TSVs) and monolithic inter-tier vias (MIVs), and compare them with GPU-only and PIM-only counterparts. Various GPU-only platforms using two main-memory technologies (DRAM, ReRAM) and three interconnect technologies (2D, TSV, MIV) are evaluated as well. Experimental results show that 3D-ReG can achieve on average 5.64× training speedup and 3.56× higher energy efficiency compared with the GPU with DRAM as main memory, at the cost of 0.05%–3.39% accuracy drop. We define a new metric, gain-loss ratio (GLR), which quantitatively evaluates the capability of a DNN training hardware in terms of the model accuracy and hardware efficiency. The results of our comparison show that the aggressive task-mapping scheme on MIV-based 3D-ReG outperforms the other methods.

In this paper, we develop physical design tools and methodologies to tackle the inter-tier performance variations caused by low temperature manufacturing in 2-tier gate-level monolithic 3D ICs (M3D). First, we model the top tier front-end-of-line (FEOL) device mobility degradation and its impact on cell delay/power values. Next, we quantify the impact of tungsten interconnect and cost-driven metal layer saving in the back-end-of-line (BEOL) of the bottom tier. These device and interconnect degradation models are used in our new full-chip M3D physical design flow named Derated 2D. This flow overcomes the well-known drawback of the state-of-the-art Shrunk 2D that requires shrinking of layout objects and RC parasitics. Also, Derated 2D performs low-temperature process-aware tier partitioning to effectively keep timing-critical components in the bottom tier. Moreover, Derated 2D conducts timing-driven monolithic inter-tier via (MIV) planning to cope with the resistivity increase in tungsten BEOL. Lastly, Derated 2D offers an effective timing closure solution through a post-route optimization. Experiments based on a foundry-grade 7nm FinFET process design kit (PDK) show that Derated 2D achieves up to 36% performance improvement and 10% energy saving compared with Shrunk 2D. Using a post-route optimization, Derated 2D further improves timing under various FEOL/BEOL degradation settings at a minimum energy overhead.